Cybersecurity is practice of preventing cyberattacks on vital infrastructure and private data. Many applications involve cybersecurity, such as in banks, hospitals, and other industry sectors, which face issues related to data insecurity, confidential information breaches and accuracy. However, high traffic processed attacks and inaccurate identification of malicious activity lead to high false positives and class imbalance. In this research, Adaptive Inertial Weight with Artificial Rabbit Optimization (AIW-ARO) is proposed for Intrusion Detection Systems (IDS) to secure entire network in cybersecurity. Initially, data is obtained from N-BaIoT dataset and then normalized using a Standard Scaler. Feature selection is performed using the AIW-ARO method to select relevant features and secure data by hiding information and using foraging as an exploration strategy, preventing detection by cybersecurity measures. Finally, classification using Long Short-Term Memory (LSTM) technique identifies attacks and classifies IDS accurately, effectively in cybersecurity. The proposed AIW-ARO technique is evaluated using the N-BaIoT datasets, achieving a higher accuracy of 99.99%. This is compared to existing techniques like Bidirectional Gated Recurrent Unit (Bi-GRU), Convolutional Neural Network (CNN), and Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM).